Abstract: Anomaly detection for time series data is one of the important research problems in data analysis.Its main challenge is to detect if there are any anomalies and locate anomalies with low delay according to context.Most of existing anomaly detection methods capture anomalies using the probability density ratio to measure similarity between sequences.These methods need to use the cross-validation method to estimate the parameters of probability density ratio.However,cross-validation can increase the computational complexity,resulting in low computational efficiency and a high time delay.To address these issues,this paper proposes a detection method based on total variation ratio separation distance,in which total variation is adopted to extract sequence fluctuation features.Due to the fact that the total variation ratio is better than probability density ratio,the proposed method achieves higher computational efficiency and lower time delay.To reduce noise interference and further improve the detection accuracy,the proposed method is combined with the relative total variation.Experimental results show that the proposed method performs well in terms of detection accuracy,low delay and computational efficiency.

Key words: Anomaly detection, Probability density ratio, Time delay, Total variation, Relative total variation